Adaptive filter-based reconstruction engine design for compressive sensing

The reconstruction in compressive sensing is an underdetermined question. Almost all existing reconstruction algorithms utilize pseudo inverse to solve this problem. However, the matrix inverse in pseudo inverse has high complexity. In this paper, we apply least mean square filter (LMS) to signal reconstruction and propose a new reconstruction algorithm for compressive sensing. The results show that proposed method has good recovery performance and low computational complexity compared with existing works. Moreover, we implemented the proposed reconstruction algorithm in 90nm CMOS which operated at 200 MHz and occupied an area of 1.36mm². Throughput of the proposed method is 70% higher than state-of-the-art under the same cost.